Systems and methods are disclosed for grouping cells in a slide image that share a similar target, comprising receiving a digital pathology image corresponding to a tissue specimen, applying a trained machine learning system to the digital pathology image, the trained machine learning system being trained to predict at least one target difference across the tissue specimen, and determining, using the trained machine learning system, one or more predicted clusters, each of the predicted clusters corresponding to a subportion of the tissue specimen associated with a target.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A computer-implemented method for grouping cells in a slide image that share a similar target, the method comprising: receiving a digital pathology image corresponding to a tissue specimen; applying a trained machine learning system to the digital pathology image, the trained machine learning system being trained to predict at least one target difference across the tissue specimen; and determining, using the trained machine learning system, one or more predicted clusters, each of the predicted clusters corresponding to a subportion of the tissue specimen associated with a target.
This invention relates to digital pathology and the automated analysis of tissue specimens. The problem addressed is the need to accurately group cells in a slide image that share similar biological or pathological characteristics, which is critical for diagnosis, research, and treatment planning. Traditional manual analysis is time-consuming and prone to human error, while existing automated methods often lack precision in identifying meaningful clusters of cells. The method involves receiving a digital pathology image of a tissue specimen and applying a trained machine learning system to analyze the image. The machine learning system is specifically trained to predict differences in biological or pathological targets across the tissue, such as protein expression levels, genetic mutations, or other cellular features. By processing the image, the system identifies one or more clusters of cells, where each cluster corresponds to a subportion of the tissue that shares a similar target characteristic. This clustering helps pathologists and researchers quickly identify regions of interest, improving diagnostic accuracy and efficiency. The method leverages advanced machine learning to automate the detection of biologically relevant patterns, reducing the need for manual intervention and enhancing reproducibility.
2. The computer-implemented method of claim 1 , further comprising: mapping, using the trained machine learning system, the predicted clusters to at least one known biological target.
This invention relates to a computer-implemented method for analyzing biological data using machine learning to predict clusters of biological entities and map them to known biological targets. The method addresses the challenge of identifying meaningful patterns in complex biological datasets, which is critical for drug discovery, disease research, and personalized medicine. The system first processes input biological data, such as gene expression profiles or protein interactions, to extract relevant features. A trained machine learning model then analyzes these features to predict clusters of biologically related entities, such as genes, proteins, or cells, that exhibit similar behavior or functions. These predicted clusters are subsequently mapped to at least one known biological target, such as a protein, gene, or molecular pathway, to provide insights into their biological significance. The mapping step helps researchers understand the functional roles of the identified clusters and their potential relevance to diseases or therapeutic interventions. By automating the clustering and mapping process, the method reduces manual effort and improves the accuracy and efficiency of biological data analysis. This approach is particularly useful in high-throughput biological research where large-scale data interpretation is required.
3. The computer-implemented method of claim 2 , further comprising: receiving at least one genetic sequence of the tissue specimen, a location of the genetic sequence being based on the mapping.
The invention relates to a computer-implemented method for analyzing tissue specimens, particularly focusing on genetic sequencing and spatial mapping. The method addresses the challenge of accurately correlating genetic data with specific locations within a tissue sample, which is critical for applications such as cancer research, diagnostics, and personalized medicine. The method involves generating a spatial map of the tissue specimen, which includes identifying and mapping distinct regions or features within the sample. This mapping provides a reference framework for subsequent genetic analysis. The method then receives at least one genetic sequence of the tissue specimen, where the location of the genetic sequence is determined based on the previously generated spatial map. This ensures that the genetic data is spatially contextualized, allowing researchers to understand how genetic variations correspond to specific regions within the tissue. By integrating spatial mapping with genetic sequencing, the method enables precise localization of genetic information, improving the accuracy of tissue analysis. This approach is particularly useful for studying heterogeneous tissues, such as tumors, where genetic variations can vary significantly across different regions. The method enhances the ability to correlate genetic mutations with their physical locations, supporting more detailed and accurate biological insights.
4. The computer-implemented method of claim 1 , further comprising: using a spatial distribution of the predicted clusters to determine, by the trained machine learning system, a recommended treatment decision.
This invention relates to a computer-implemented method for analyzing medical data to improve treatment decisions. The method addresses the challenge of optimizing patient outcomes by leveraging machine learning to predict clusters of medical conditions and using their spatial distribution to recommend treatments. The method involves training a machine learning system on medical data to predict clusters of related conditions. These clusters are analyzed to identify patterns in their spatial distribution, which may indicate underlying disease progression or treatment effectiveness. The trained system then uses this spatial distribution to determine a recommended treatment decision, ensuring that the chosen intervention aligns with the observed patterns in patient data. The invention builds on a broader method that includes collecting medical data, preprocessing it, and training a machine learning model to predict clusters of conditions. The spatial distribution analysis enhances this process by incorporating geographic or anatomical relationships between clusters, allowing for more informed treatment recommendations. This approach improves decision-making by considering both the predicted clusters and their spatial context, leading to more personalized and effective treatments.
5. The computer-implemented method of claim 1 , wherein applying the trained machine learning system comprises using an artificial intelligence (AI)-predicted segmentation with a clustering heuristic to determine an optimal sampling location of the tissue specimen to maximize information gained about the target difference across the tissue specimen.
This invention relates to a computer-implemented method for analyzing tissue specimens using machine learning to optimize sampling locations. The method addresses the challenge of efficiently identifying regions of interest in tissue samples to maximize the information gained about specific biological differences, such as disease markers or structural variations. The system leverages a trained machine learning model to predict segmentation of the tissue, which is then refined using a clustering heuristic. This approach ensures that the selected sampling locations provide the most relevant data, reducing the need for exhaustive manual analysis. The clustering heuristic groups similar regions within the tissue, allowing the system to prioritize areas with the highest potential for meaningful insights. By combining AI-driven segmentation with clustering, the method enhances the accuracy and efficiency of tissue analysis, particularly in medical diagnostics and research applications. The technique is applicable to various imaging modalities, including histopathology slides and other biological samples, where precise sampling is critical for accurate diagnosis or study. The overall goal is to streamline the analysis process while maximizing the informational yield from limited tissue samples.
6. The computer-implemented method of claim 1 , further comprising: determining a target composition of a set of samples associated with the tissue specimen, the samples having been processed using an available sequencing technique; and using an AI-predicted segmentation to infer a spatial distribution over factors in a slide.
This invention relates to computational pathology, specifically improving the analysis of tissue specimens using sequencing techniques and artificial intelligence (AI). The problem addressed is the challenge of accurately determining the spatial distribution of biological factors within a tissue sample, which is critical for understanding disease mechanisms and developing targeted therapies. Traditional sequencing methods often lack spatial resolution, making it difficult to correlate molecular data with tissue structure. The method involves processing a set of samples from a tissue specimen using an available sequencing technique, such as RNA sequencing or DNA sequencing. The samples are analyzed to determine their composition, which may include molecular markers, gene expressions, or other biological factors. An AI-predicted segmentation is then used to infer the spatial distribution of these factors across the tissue slide. The AI model likely leverages machine learning techniques to segment the tissue into regions of interest, such as tumor cells, healthy cells, or different tissue types, based on the sequencing data and possibly additional imaging data. By combining sequencing data with AI-driven segmentation, the method provides a more comprehensive understanding of the spatial organization of biological factors within the tissue. This can enhance diagnostic accuracy, improve treatment planning, and support research into disease progression. The approach may also help identify subpopulations of cells within a tissue that exhibit distinct molecular profiles, which is valuable for precision medicine.
7. The computer-implemented method of claim 1 , the trained machine learning system having been trained by: receiving a plurality of digital pathology training images, the training images containing or being associated with tissue metadata; and training the trained machine learning system to predict clusters based on the digital pathology training images and the tissue metadata.
This invention relates to a computer-implemented method for training a machine learning system to analyze digital pathology images. The method addresses the challenge of accurately classifying or clustering tissue samples in pathology by leveraging both image data and associated metadata. The system is trained using a dataset of digital pathology training images, each of which contains or is linked to tissue metadata. The training process involves processing these images and metadata to enable the machine learning system to predict clusters or classifications of tissue samples. The system learns to identify patterns and relationships between the visual features in the images and the metadata, improving its ability to categorize tissue samples for diagnostic or research purposes. This approach enhances the accuracy and reliability of automated pathology analysis, reducing the need for manual review and improving efficiency in medical diagnostics. The trained system can be applied to new, unlabeled pathology images to predict their clusters or classifications based on the learned patterns.
8. The computer-implemented method of claim 7 , wherein training the trained machine learning system further comprises: using at least one of supervised learning, weakly supervised learning, and/or unsupervised learning.
This invention relates to machine learning systems and methods for training such systems using different learning paradigms. The problem addressed is the need for flexible and adaptable training approaches in machine learning to handle varying data quality, labeling availability, and task requirements. The invention provides a method for training a machine learning system that can employ supervised learning, weakly supervised learning, or unsupervised learning, or a combination thereof. Supervised learning involves training the system using labeled data where input-output pairs are explicitly provided. Weakly supervised learning uses data with noisy or incomplete labels, allowing the system to learn from imperfect annotations. Unsupervised learning enables the system to identify patterns and structures in unlabeled data, making it useful when labeled data is scarce or unavailable. The method allows the selection of one or more of these learning approaches based on the specific requirements of the task, the nature of the available data, and the desired performance characteristics of the trained system. This flexibility enhances the applicability of the machine learning system across different domains and scenarios, improving its robustness and adaptability. The invention also includes techniques for integrating these learning paradigms to optimize training efficiency and accuracy.
9. The computer-implemented method of claim 8 , wherein using supervised learning comprises: inputting the plurality of digital pathology training images of tissues with a spatial transcriptomic and/or imaging mass cytometry techniques; and training the trained machine learning system for segmentation of predicting a ground truth spatial transcriptomic and/or genetic marker masks.
This invention relates to a computer-implemented method for analyzing digital pathology images using supervised learning to enhance spatial transcriptomic and imaging mass cytometry analysis. The method addresses the challenge of accurately segmenting and predicting spatial transcriptomic or genetic marker masks from digital pathology images, which is critical for understanding tissue composition and disease mechanisms. The method involves inputting a plurality of digital pathology training images of tissues, where these images are generated using spatial transcriptomic techniques or imaging mass cytometry. These techniques provide high-resolution data on gene expression and protein distribution within tissue samples. The method then trains a machine learning system to segment and predict ground truth spatial transcriptomic or genetic marker masks. The trained system can accurately identify and delineate regions of interest within the tissue images, such as different cell types or molecular markers, based on the training data. The supervised learning approach ensures that the machine learning model learns from labeled training data, improving its ability to generalize to new, unseen pathology images. This method enhances the precision of spatial transcriptomic and imaging mass cytometry analysis, enabling more accurate and reliable insights into tissue biology and pathology. The trained system can be applied to various diagnostic and research applications, improving the efficiency and accuracy of tissue analysis in clinical and research settings.
10. The computer-implemented method of claim 8 , wherein using weakly supervised learning comprises: using knowledge about a presence or absence of at least one target in a slide; and applying Multiple Instance Learning to learn a segmentation or predicting the at least one target spatially.
This invention relates to computer-implemented methods for analyzing biological or medical slides using weakly supervised learning techniques. The problem addressed is the difficulty of accurately identifying and segmenting targets (such as cells, tissues, or other structures) in slides when only limited labeled data is available. Traditional supervised learning requires extensive manual annotation, which is time-consuming and impractical for large datasets. The method leverages weakly supervised learning, specifically Multiple Instance Learning (MIL), to overcome this limitation. It uses prior knowledge about the presence or absence of at least one target within a slide, rather than requiring precise pixel-level annotations. By applying MIL, the system learns to predict the spatial location of the target within the slide. This approach allows the model to generalize from coarse labels (e.g., slide-level labels) to fine-grained spatial predictions, improving efficiency and accuracy in target detection and segmentation. The method may involve preprocessing the slide, extracting features, and training a model to classify instances (e.g., image patches) while considering the weak labels. The trained model can then predict the spatial distribution of the target, enabling applications in pathology, diagnostics, or biological research where detailed annotations are scarce. This technique reduces reliance on expert annotation while maintaining robust performance.
11. The computer-implemented method of claim 8 , wherein unsupervised learning comprises: using autoencoders to learn and generate high-level embeddings for small regions of slides that encode morphological differences; and using k-means and/or hierarchical clustering to generate groupings of cells with similar morphological embeddings.
This invention relates to digital pathology and automated cell classification in histological slides. The method addresses the challenge of accurately identifying and grouping cells with similar morphological features in large-scale tissue images, which is critical for disease diagnosis and research. The approach leverages unsupervised learning techniques to analyze small regions of slides without requiring pre-labeled data. The method uses autoencoders to extract high-level embeddings from these regions, capturing subtle morphological variations. These embeddings are then processed using clustering algorithms such as k-means or hierarchical clustering to group cells with similar morphological characteristics. This automated classification reduces manual effort and improves consistency in pathological analysis. The technique is particularly useful for identifying cell types, detecting abnormalities, or quantifying tissue composition in digital pathology workflows. By eliminating the need for manual annotation, it accelerates research and clinical applications where large datasets must be analyzed efficiently. The combination of autoencoders and clustering enables scalable, data-driven morphological analysis in histological imaging.
12. The computer-implemented method of claim 1 , further comprising using flow cytometry and mass spectrometry techniques to determine physical and chemical characteristics of the tissue specimen.
This invention relates to a method for analyzing tissue specimens using advanced analytical techniques to determine their physical and chemical properties. The method involves preparing a tissue specimen for analysis, which may include sectioning, staining, or other preparatory steps to enhance the visibility or detectability of specific features. The prepared specimen is then subjected to flow cytometry, a technique that measures the physical and chemical characteristics of cells or particles in a fluid stream by passing them through a laser beam and analyzing the scattered light and fluorescence. Additionally, mass spectrometry is employed to identify and quantify the molecular composition of the specimen, providing detailed chemical information. The combination of these techniques allows for comprehensive characterization of the tissue, including cellular morphology, protein expression, and metabolic profiles. This approach is particularly useful in biomedical research, diagnostics, and personalized medicine, where detailed tissue analysis is critical for understanding disease mechanisms or developing targeted therapies. The method may also include data processing steps to integrate and interpret the results from both flow cytometry and mass spectrometry, enabling more accurate and efficient analysis of tissue specimens.
13. The computer-implemented method of claim 1 , further comprising outputting an image identifying the predicted clusters as one or more pixel masks.
This invention relates to computer vision and clustering techniques for image analysis. The method addresses the challenge of identifying and visualizing distinct regions or clusters within an image, which is useful in applications like object detection, segmentation, and medical imaging. The method processes an input image to generate predicted clusters representing different regions of interest. These clusters are then converted into one or more pixel masks, where each mask highlights a specific cluster within the image. The pixel masks provide a clear visual representation of the detected regions, enabling further analysis or decision-making. The method may involve preprocessing the image, applying clustering algorithms, and refining the cluster boundaries to improve accuracy. The output pixel masks can be used in various downstream tasks, such as image annotation, automated quality control, or medical diagnosis. The invention enhances the interpretability and usability of clustering results in image processing workflows.
14. The computer-implemented method of claim 1 , wherein the predicted clusters comprise a set of points corresponding to a center of a found antigen cluster and/or a pixel mask segmenting an image into one or more antigen clusters.
This invention relates to image analysis in biological or medical imaging, specifically for identifying and segmenting antigen clusters in microscopic images. The method addresses the challenge of accurately detecting and delineating antigen clusters, which is critical for applications such as immunohistochemistry (IHC) analysis, where precise identification of protein expression patterns is essential for diagnosis and research. The method involves processing an image to predict clusters of antigens, where each cluster is represented by either a set of points corresponding to the center of a detected antigen cluster or a pixel mask that segments the image into distinct antigen clusters. The segmentation ensures that overlapping or closely positioned clusters are accurately separated, improving the reliability of downstream analysis. The technique leverages computational algorithms, likely machine learning or image processing, to enhance the precision of antigen detection compared to manual or less sophisticated automated methods. By providing both center-point and mask-based representations, the method offers flexibility in how the clusters are utilized, accommodating different analytical workflows. This approach is particularly valuable in pathology, where accurate antigen localization is crucial for diagnosing conditions like cancer, where protein expression patterns can indicate disease progression or treatment response. The invention improves upon prior methods by offering a more robust and adaptable solution for antigen cluster identification in digital pathology.
15. The computer-implemented method of claim 1 , further comprising: determining predicted mappings using the predicted clusters, the predicted mappings comprising a map from any of at least one cluster, of the predicted clusters, to a predicted antigen, and for each of the predicted clusters, a plurality of likely antigens.
This invention relates to a computer-implemented method for predicting antigen mappings from biological data, particularly in the context of immunology or bioinformatics. The method addresses the challenge of accurately identifying which antigens are associated with specific clusters of biological data, such as immune receptor sequences or other molecular markers. The core process involves generating predicted clusters from input data, which may include sequencing data or other biological measurements. These clusters represent groups of related biological entities, such as immune cells or receptor sequences, that share functional or structural similarities. The method then determines predicted mappings by associating these clusters with specific antigens. Each cluster is mapped to at least one predicted antigen, and for each cluster, multiple likely antigens are identified, reflecting the probabilistic nature of antigen recognition. This approach helps in understanding immune responses, designing vaccines, or developing diagnostic tools by providing a structured way to link biological data to potential antigens. The method leverages computational techniques to handle large datasets and improve the accuracy of antigen predictions, addressing limitations in traditional experimental approaches.
16. A system for grouping cells in a slide image that share a similar target, comprising: at least one memory storing instructions; and at least one processor configured to execute the instructions to perform operations comprising: receiving a digital pathology image corresponding to a tissue specimen; applying a trained machine learning system to the digital pathology image, the trained machine learning system being trained to predict at least one target difference across the tissue specimen; and determining, using the trained machine learning system, one or more predicted clusters, each of the predicted clusters corresponding to a subportion of the tissue specimen associated with a target.
The system is designed for grouping cells in digital pathology images based on shared biological or morphological characteristics. In pathology, analyzing tissue samples often requires identifying regions with similar properties, such as protein expression levels, cell types, or disease markers. Manual analysis is time-consuming and prone to human error, making automated clustering essential for efficiency and accuracy. The system processes a digital pathology image of a tissue specimen using a trained machine learning model. The model predicts variations in a target attribute across the tissue, such as protein expression or cell morphology. By analyzing these predictions, the system identifies clusters of cells that share similar target values. Each cluster corresponds to a distinct subregion of the tissue, enabling pathologists to study specific areas of interest. The machine learning model is pre-trained to recognize patterns in tissue images and correlate them with the target attribute. The system automates the clustering process, reducing manual effort and improving consistency in pathological analysis. This approach enhances diagnostic accuracy and supports research by providing structured data for further investigation. The system is applicable in clinical diagnostics, drug development, and biomedical research where precise tissue characterization is critical.
17. The system of claim 16 , the operations further comprising: mapping, using the trained machine learning system, the predicted clusters to at least one known biological target.
The invention relates to a machine learning-based system for analyzing biological data to identify and map predicted clusters to known biological targets. The system processes input data, such as molecular or genetic information, to generate clusters representing distinct biological patterns or groupings. A trained machine learning model predicts these clusters, which are then mapped to at least one known biological target, such as a protein, gene, or cellular pathway. This mapping helps identify potential biological mechanisms or therapeutic targets. The system may also include preprocessing steps to prepare the input data, feature extraction to highlight relevant biological characteristics, and validation techniques to ensure the accuracy of the predicted clusters. The machine learning model is trained using labeled biological data, allowing it to recognize patterns and relationships within the input data. By mapping predicted clusters to known biological targets, the system facilitates research in drug discovery, disease mechanisms, and personalized medicine. The invention improves upon existing methods by automating the identification and mapping process, reducing manual effort and increasing efficiency in biological data analysis.
18. The system of claim 17 , the operations further comprising: receiving at least one genetic sequence of the tissue specimen, a location of the genetic sequence being based on the mapping.
The system is designed for analyzing tissue specimens, particularly for identifying and mapping genetic sequences within the tissue. The technology addresses the challenge of accurately locating and interpreting genetic information in biological samples, which is critical for applications such as disease diagnosis, genetic research, and personalized medicine. The system generates a spatial map of the tissue specimen, which includes positional data for various regions of the tissue. This mapping allows for precise identification of where specific genetic sequences are located within the tissue. The system then receives at least one genetic sequence of the tissue specimen, with the location of the genetic sequence being determined based on the spatial mapping. This integration of genetic sequence data with spatial information enables more accurate analysis of the tissue's genetic composition and its spatial distribution. The system may also include additional features such as imaging the tissue specimen, processing the image to identify regions of interest, and correlating the genetic sequence data with the spatial map to provide a comprehensive understanding of the tissue's genetic landscape. This approach enhances the ability to study genetic variations, mutations, or other relevant biological markers in their spatial context, improving diagnostic accuracy and research capabilities.
19. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform operations for grouping cells in a slide image that share a similar target, the operations comprising: receiving a digital pathology image corresponding to a tissue specimen; applying a trained machine learning system to the digital pathology image, the trained machine learning system being trained to predict at least one target difference across the tissue specimen; and determining, using the trained machine learning system, one or more predicted clusters, each of the predicted clusters corresponding to a subportion of the tissue specimen associated with a target.
The invention relates to digital pathology and the automated analysis of tissue specimens in slide images. The problem addressed is the need to accurately group cells or regions within a tissue specimen that share similar biological or pathological characteristics, such as protein expression levels, genetic mutations, or other diagnostic markers. Traditional manual analysis is time-consuming and prone to human error, while existing computational methods often lack precision in identifying meaningful clusters. The solution involves a machine learning-based approach for segmenting and clustering cells or regions in a digital pathology image. A trained machine learning system is applied to the image to predict differences in a target property (e.g., protein expression, cell type, or mutation status) across the tissue specimen. The system then identifies clusters of cells or regions that exhibit similar target properties, effectively grouping them into subportions of the tissue. This allows pathologists to quickly identify areas of interest for further analysis, improving diagnostic accuracy and efficiency. The method leverages pre-trained models to automate the clustering process, reducing manual effort and enhancing reproducibility. The output is a segmented image where clusters are visually distinguishable, aiding in pathological assessment.
20. The non-transitory computer-readable medium of claim 19 , the operations further comprising: mapping, using the trained machine learning system, the predicted clusters to at least one known biological target.
This invention relates to a machine learning system for analyzing biological data to identify and map predicted clusters to known biological targets. The system processes input data, such as molecular or genomic information, to generate clusters representing groups of related biological entities. A trained machine learning model predicts these clusters, which are then mapped to at least one known biological target, such as a protein, gene, or cellular pathway. The mapping step involves correlating the predicted clusters with existing biological knowledge to identify functional relationships or therapeutic relevance. The system may also include preprocessing steps to prepare the input data, such as normalization or feature extraction, and post-processing to refine the predicted clusters before mapping. The machine learning model is trained using labeled biological data to improve accuracy in cluster prediction and target mapping. This approach enables researchers to discover novel biological insights or validate hypotheses by linking computational predictions to established biological knowledge. The invention is particularly useful in drug discovery, genomics, and systems biology, where understanding the relationships between molecular entities and biological functions is critical.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
August 2, 2021
April 19, 2022
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.